feat(transport): congestion control abstraction layer (trait-based CC strategy)

[Yanstart]

Context

Quick- currently selects CC via a simple enum (Cubic | Bbr | Bbr2) mapped to quiche's one-time config.set_cc_algorithm() call. This works for a single-backend architecture but limits future evolution.

Goal

Create a trait-based CC abstraction that decouples Quick-'s transport layer from quiche's internal CC implementation. This enables:

• Pluggable CC backends (not just quiche-internal algorithms)
• Runtime CC comparison/A/B testing framework
• Future multipath-QUIC (feat(transport): Multipath-QUIC preparation #49) where different paths may use different CC
• Potential integration of non-quiche QUIC backends

Requirements

• CongestionController trait abstracting CC selection, parameterization, and metrics
• Benchmark framework for A/B CC comparison on identical network conditions
• Preserve current enum simplicity for the common case (quiche-internal CC)
• Zero overhead when using quiche's built-in CC (trait should compile away)

Dependencies

• feat(transport): BBRv3 readiness and abstraction layer #48 (BBRv3 implementation — informs what the trait needs to expose)
• Fork quiche: satellite-optimized recovery constants (packet_threshold, ACK frequency, PTO) #36 (quiche fork — current CC parameter exposure)

Non-goals

• Not replacing quiche's internal CC — wrapping it
• Not implementing CC algorithms in Rust outside quiche